为了监测黄瓜采摘后贮藏期间品质的变化, 借助不同贮藏日期贮藏室气氛的3D荧光数据, 提出了一种特征荧光信息(特征激发与特征发射波长)提取方法, 实现了黄瓜贮藏期间品质的监测。 首先, 对3D荧光数据进行去除瑞利散射和多项式Savitzky-Golar(SG)平滑降噪预处理, 有效消除了散射和噪声信号的影响。 其次, 对预处理后的3D荧光数据进行主成分分析(PCA)得到主成分矩阵, 并运用各主成分变量构造Wilks统计量, 选取了最小值对应的主成分(第11主成分, PC11); 根据构造该主成分的各原始变量(激发波长)的组合系数大小提取了8个特征激发波长。 然后, 采用10 nm的间隔对发射光谱进行了波段划分, 运用小波包分解(WPD)对每个波段进行了3尺度分解, 计算了各波段分解后的小波包能量, 综合8天试验结果选择能量最高的发射波段作为初选发射波段。 采用偏最小二乘回归(PLS)结合黄瓜理化指标(硬度、 叶绿素含量和失重率)对初选的发射波段进行了分析, 依据回归系数精选了7个特征发射波长, 简化了计算。 同时, 根据黄瓜硬度数据初步找到了其变化趋势的转折点; 根据黄瓜叶绿素含量变化曲线及一阶导数, 发现了叶绿素下降趋势最显著的点, 并结合试验过程中的感官观察结果, 确定第5个贮藏日为黄瓜品质突变日, 并选择第5个贮藏日为监测基准日。 最后, 采用提取的特征荧光信息计算不同贮藏天数与监测基准日之间的马氏距离(MD), 构建MD监测模型。 结果表明, 随着贮藏时间越来越接近监测基准日, MD值则逐渐减小到0, 与黄瓜贮藏过程中品质变化进程相符。 上述多变量统计分析融合小波包能量的特征波长提取方法和应用特征荧光信息构建的MD监测模型有望成为黄瓜贮藏过程中品质监测的一种可行方法。

To monitor the quality change of cucumber during storage, a feature extraction method of 3D fluorescence information of storage room gas (feature excitation wavelength and feature emission wavelength) is proposed using multivariate statistical analysis coupled with wavelet packet energy during different cucumber storage dates. Firstly, these 3D fluorescence data were handled by removing Rayleigh scattering and polynomial Savitzky-Golar (SG) smoothing to remove the effects of scattering and noise signals. Secondly, the pre-processed 3D fluorescence data were handled by principal component analysis (PCA) to obtain the principal component matrix, and Wilks statistics were constructed by using each principal component variable, then the principal component corresponding to the minimum value (the 11th principal component, PC11) was selected. Then eight feature excitation wavelengths were extracted according to the combination coefficient of each original variable (excitation wavelength) of the principal component. Thirdly, the emission spectrum is divided by 10nm interval and gets 26 bands; the 3-scale based wavelet packet decomposition (WPD) was carried out for each band, and the wavelet packet energy of each band after decomposition was calculated. And then, according to the analysis results of 8 days test data, the band with the highest energy was selected as the primary feature emission band. Fourthly, partial least squares regression (PLS) was used to analyze the primary emission bands combined with the physicochemical indexes (hardness, chlorophyll content and weight loss rate) of cucumber, and seven feature emission wavelengths were selected according to the regression coefficient, which greatly simplified the calculation. At the same time, according to the hardness data of cucumber, the turning point of its trend change could be found; and according to the cucumber chlorophyll content data and its first derivative, the chlorophyll decline the most significant points could be also found; and then combined with the sensory analysis in the process of test, ultimately determine the quality of cucumber stored at the fifth day become bad rapidly. Therefore, the fifth storage day was chosen to monitor the reference date. Finally, Mahalanobis Distance (MD) between different storage days and monitoring reference dates was calculated using the extracted feature fluorescence information, and the MD monitoring model was constructed. The results show that the MD decreased gradually to 0 with the storage time approaching the monitoring reference date, which was consistent with the quality change process of cucumber during storage. Therefore, the above feature wavelength extraction method based on multivariate statistical analysis combined with wavelet packet energy and the MD monitoring model is expected to be an effective method for quality monitoring of cucumber during storage.